# v17d interpretation audit

Frozen status is `v17d_endpoint_centers_not_stable`. All traversal, resource, reversibility, representation, residual-profile and proposal-footprint layers passed, but endpoint centers passed only `85/108` and endpoint-distance agreement `12/18`. Every distance failure was the start-family contrast; seed and early/late distance comparisons passed `12/12`.

The descriptive postrun audit found source-edge/conflict start-gap contraction in `12/12` cells but essentially flat direct cross-start distance. Exact residual-SCC profile identity across the eight representative endpoints per source rules out a changing residual partition as the explanation, not hidden bounded-cycle state-graph separation or slow traversal.

This gate supports only finite effect-blind diagnostics on six reused state spaces. It does not prove irreducibility, convergence, mixing time, global uniformity or a canonical measure. Residual-SCC and accepted-footprint overlap must not be renamed global Markov components. Source spectrum and observed-effect computations remained prohibited.